Deep Cogito, a startup building models that improve themselves, has raised $43M in a Series A led by TQ Ventures, with Benchmark, Nexus Venture Partners, Atreides Management, South Park Commons and Zscaler participating. Total outside funding tops $56M.
Founded in 2024 by former Google engineers Drishan Arora and Dhruv Malrana, the company develops the Cogito family of open-source large language models. Its training uses an internal technique called IDA, which expands the infrastructure available to a model during a prompt and analyzes the resulting quality gain to refine parameters. A process supervision method grades each reasoning step, cutting unnecessary moves and lowering hardware use.
Deep Cogito monetizes through a platform that lets enterprises build custom models on internal data, with investor Zscaler among its users. The fresh capital will expand the enterprise business, grow the research team and fund new open-source releases.
It is one of several startups chasing self-improving AI, a field that has drawn multibillion-dollar checks from investors betting that models that can train themselves will reshape the industry.
